Current trend in data centers is a growing awareness on energy consumption and a search for energy efficient solutions that keep in view user required performance and various service level agreements (SLA). Efficient resource management policies, in particular, can significantly reduce the energy consumption, of the data centers. In existing data centers virtual machine (VM) technology is being used as a unit of resource provisioning. VMs offer flexibilities such as preemption, migration and consolidation in order to implement energy efficient policies.
Further, the data centers support a combination of advance-reservation and best-effort schemes, interactive and batch jobs, tight deadline and loose-deadline jobs. When a shortage of resources such as the virtual machines, exist, commonly, lower priority requests are preempted in favor of a higher priority request. For instance, Haizea, a resource scheduler for virtualized data centers, supports a combination of advance-reservation (AR) and best-effort (BE) requests. In the former, the resources must be available at the requested time, whereas in the latter, requests can be served as soon as possible and are placed in a queue if necessary. The BE requests are usually preempted when an insufficiency of resources exists, for a newly arriving AR request. However, preemption may lead to starvation of the BE requests. In order to prevent such starvation, a limited and predictable waiting time for BE requests, needs to be maintained and monitored, such that the waiting time is not violated for any BE request.
Hence, there is a need for a method and a system that can add a power-awareness capability to the Haizea such that, the arriving AR request is served in manner that shall avoid starvation of the BE requests. The alternate method and system, must preempt the BE requests for serving the arriving AR request, reactivate switched off resources, and alternatively perform VM consolidation to save energy in circumstances that shall avoid starvation of the BE requests. Thus a method and a system for dynamic provisioning of resources in a virtualized data center is proposed.